Add historical based algorithm

client/internal/lazyconn/inactivity/manager.go

@@ -1,7 +1,12 @@
 package inactivity
 
 import (
+	"container/list"
 	"context"
+	"fmt"
+	"math"
+	"os"
+	"strconv"
 	"time"
 
 	log "github.com/sirupsen/logrus"
@@ -24,40 +29,86 @@ const (
 	handshakeRespBytes   = 92
 	handshakeMaxInterval = 3 * time.Minute
 
-	checkInterval        = keepAliveInterval // todo: 5 * time.Second
+	checkInterval = 1 * time.Minute
-	keepAliveCheckPeriod = keepAliveInterval
 
-	inactivityThreshold = 3 // number of checks to consider peer inactive
+	DefaultInactivityThreshold = 5 * time.Minute
-)
-
-const (
-	// todo make it configurable
-	DefaultInactivityThreshold = 60 * time.Minute // idle after 1 hour inactivity
 	MinimumInactivityThreshold = 3 * time.Minute
+
+	recorderEnv = "NB_LAZYCONN_RECORDER_ENABLED"
 )
 
 type WgInterface interface {
 	GetStats() (map[string]configurer.WGStats, error)
 }
 
-type peerInfo struct {
+type peerHistory struct {
-	lastRxBytesAtLastIdleCheck int64
+	lastRxBytes     int64      // last received bytes
-	lastIdleCheckAt            time.Time
+	bytesHistory    *list.List // linked list of int64
-	inActivityInRow            int
+	historySize     int
+	summarizedBytes int64
 	log             *log.Entry
 }
 
+func newPeerHistory(log *log.Entry, historySize int) *peerHistory {
+	return &peerHistory{
+		bytesHistory: list.New(),
+		historySize:  historySize,
+		log:          log,
+	}
+}
+
+func (pi *peerHistory) appendRxBytes(rxBytes int64) {
+	// If at capacity, remove the oldest element (front)
+	if pi.bytesHistory.Len() == pi.historySize {
+		pi.summarizedBytes -= pi.bytesHistory.Front().Value.(int64)
+		pi.bytesHistory.Remove(pi.bytesHistory.Front())
+	}
+
+	// Add the new rxBytes at the back
+	pi.bytesHistory.PushBack(rxBytes)
+	pi.summarizedBytes += rxBytes
+}
+
+func (pi *peerHistory) historyString() string {
+	var history []string
+	for e := pi.bytesHistory.Front(); e != nil; e = e.Next() {
+		history = append(history, fmt.Sprintf("%d", e.Value.(int64)))
+	}
+	return fmt.Sprintf("%s", history)
+}
+
+func (pi *peerHistory) reset() {
+	for e := pi.bytesHistory.Front(); e != nil; e = e.Next() {
+		e.Value = int64(0)
+	}
+	pi.summarizedBytes = 0
+}
+
 type Manager struct {
 	InactivePeersChan chan []string
 	iface             WgInterface
-	interestedPeers   map[string]*peerInfo
+	interestedPeers   map[string]*peerHistory
+
+	maxBytesPerPeriod int64
+	historySize       int // Size of the history buffer for each peer, used to track received bytes over time
+	recorder          *Recorder
 }
 
-func NewManager(iface WgInterface) *Manager {
+func NewManager(iface WgInterface, configuredThreshold *time.Duration) *Manager {
+	inactivityThreshold, err := validateInactivityThreshold(configuredThreshold)
+	if err != nil {
+		inactivityThreshold = DefaultInactivityThreshold
+		log.Warnf("invalid inactivity threshold configured: %v, using default: %v", err, DefaultInactivityThreshold)
+	}
+
+	expectedMaxBytes := calculateExpectedMaxBytes(inactivityThreshold)
+	log.Infof("receive less than %d bytes per %v, will be considered inactive", expectedMaxBytes, inactivityThreshold)
 	return &Manager{
 		InactivePeersChan: make(chan []string, 1),
 		iface:             iface,
-		interestedPeers:   make(map[string]*peerInfo),
+		interestedPeers:   make(map[string]*peerHistory),
+		historySize:       calculateHistorySize(inactivityThreshold),
+		maxBytesPerPeriod: expectedMaxBytes,
 	}
 }
 
@@ -67,9 +118,7 @@ func (m *Manager) AddPeer(peerCfg *lazyconn.PeerConfig) {
 	}
 
 	peerCfg.Log.Debugf("adding peer to inactivity manager")
-	m.interestedPeers[peerCfg.PublicKey] = &peerInfo{
+	m.interestedPeers[peerCfg.PublicKey] = newPeerHistory(peerCfg.Log, m.historySize)
-		log: peerCfg.Log,
-	}
 }
 
 func (m *Manager) RemovePeer(peer string) {
@@ -83,6 +132,12 @@ func (m *Manager) RemovePeer(peer string) {
 }
 
 func (m *Manager) Start(ctx context.Context) {
+	enabled, err := strconv.ParseBool(os.Getenv(recorderEnv))
+	if err == nil && enabled {
+		m.recorder = NewRecorder()
+		defer m.recorder.Close()
+	}
+
 	ticker := newTicker(checkInterval)
 	defer ticker.Stop()
 
@@ -124,76 +179,69 @@ func (m *Manager) checkStats(now time.Time) ([]string, error) {
 
 	var idlePeers []string
 
-	for peer, info := range m.interestedPeers {
+	for peer, history := range m.interestedPeers {
 		stat, found := stats[peer]
 		if !found {
 			// when peer is in connecting state
-			info.log.Warnf("peer not found in wg stats")
+			history.log.Warnf("peer not found in wg stats")
 		}
 
-		// First measurement: initialize
+		deltaRx := stat.RxBytes - history.lastRxBytes
-		if info.lastIdleCheckAt.IsZero() {
+		if deltaRx < 0 {
-			info.lastIdleCheckAt = now
+			deltaRx = 0 // reset to zero if negative
-			info.lastRxBytesAtLastIdleCheck = stat.RxBytes
+			history.reset()
-			info.log.Infof("initializing RxBytes: %v, %v", now, stat.RxBytes)
+		}
+
+		m.recorder.ReceivedBytes(peer, now, deltaRx)
+
+		history.lastRxBytes = stat.RxBytes
+		history.appendRxBytes(deltaRx)
+
+		// not enough history to determine inactivity
+		if history.bytesHistory.Len() < m.historySize {
+			history.log.Tracef("not enough history to determine inactivity, current history size: %d, required: %d", history.bytesHistory.Len(), m.historySize)
 			continue
 		}
 
-		// check only every idleCheckDuration
+		history.log.Tracef("summarized Bytes: %d", history.summarizedBytes)
-		if shouldSkipIdleCheck(now, info.lastIdleCheckAt) {
+		if history.summarizedBytes <= m.maxBytesPerPeriod {
-			continue
-		}
-
-		// sometimes we measure false inactivity, so we need to check if we have inactivity in a row
-		if isInactive(stat, info) {
-			info.inActivityInRow++
-		} else {
-			info.inActivityInRow = 0
-		}
-
-		info.log.Infof("peer inactivity counter: %d", info.inActivityInRow)
-		if info.inActivityInRow >= inactivityThreshold {
-			info.log.Infof("peer is inactive for %d checks, marking as inactive", info.inActivityInRow)
 			idlePeers = append(idlePeers, peer)
-			info.inActivityInRow = 0
+			history.log.Tracef("peer is inactive, summarizedBytes: %d, maxBytesPerPeriod: %d, %v", history.summarizedBytes, m.maxBytesPerPeriod, history.historyString())
+		} else {
+			history.log.Tracef("peer is active, summarizedBytes: %d, maxBytesPerPeriod: %d, %v", history.summarizedBytes, m.maxBytesPerPeriod, history.historyString())
 		}
-		info.lastIdleCheckAt = now
-		info.lastRxBytesAtLastIdleCheck = stat.RxBytes
 	}
 
 	return idlePeers, nil
 }
 
-func isInactive(stat configurer.WGStats, info *peerInfo) bool {
+func validateInactivityThreshold(configuredThreshold *time.Duration) (time.Duration, error) {
-	rxSyncPrevPeriod := stat.RxBytes - info.lastRxBytesAtLastIdleCheck
+	if configuredThreshold == nil {
-
+		return DefaultInactivityThreshold, nil
-	// when the peer reconnected we lose the rx bytes from between the reset and the last check.
+	}
-	// We will suppose the peer was active
+	if *configuredThreshold < MinimumInactivityThreshold {
-	if rxSyncPrevPeriod < 0 {
+		return 0, fmt.Errorf("configured inactivity threshold %v is too low, using %v", *configuredThreshold, MinimumInactivityThreshold)
-		info.log.Debugf("rxBytes decreased, resetting last rxBytes at last idle check")
+	}
-		return false
+	return *configuredThreshold, nil
 }
 
-	switch rxSyncPrevPeriod {
+// calculateHistorySize calculates the number of history entries needed based on the inactivity threshold.
-	case 0:
+func calculateHistorySize(inactivityThreshold time.Duration) int {
-		info.log.Debugf("peer inactive, received 0 bytes")
+	return int(math.Ceil(inactivityThreshold.Minutes() / checkInterval.Minutes()))
-		return true
-	case keepAliveBytes:
-		info.log.Debugf("peer inactive, only keep alive received, current RxBytes: %d", rxSyncPrevPeriod)
-		return true
-	case handshakeInitBytes + keepAliveBytes:
-		info.log.Debugf("peer inactive, only handshakeInitBytes + keepAliveBytes, current RxBytes: %d", rxSyncPrevPeriod)
-		return true
-	case handshakeRespBytes + keepAliveBytes:
-		info.log.Debugf("peer inactive, only handshakeRespBytes + keepAliveBytes, current RxBytes: %d", rxSyncPrevPeriod)
-		return true
-	default:
-		info.log.Infof("active, RxBytes: %d", rxSyncPrevPeriod)
-		return false
-	}
 }
 
-func shouldSkipIdleCheck(now, lastIdleCheckAt time.Time) bool {
+func calculateExpectedMaxBytes(duration time.Duration) int64 {
-	minDuration := keepAliveCheckPeriod - (checkInterval / 2)
+	// Calculate number of keep-alive packets expected
-	return now.Sub(lastIdleCheckAt) < minDuration
+	keepAliveCount := int64(duration.Seconds() / keepAliveInterval.Seconds())
+	keepAliveBytes := keepAliveCount * keepAliveBytes
+
+	// Calculate potential handshake packets (conservative estimate)
+	handshakeCount := int64(duration.Minutes() / handshakeMaxInterval.Minutes())
+	if handshakeCount == 0 && duration >= handshakeMaxInterval {
+		handshakeCount = 1
+	}
+	handshakeBytes := handshakeCount * (handshakeInitBytes + keepAliveBytes) // handshake + extra bytes
+
+	// todo: fine tune this value, add some overhead for unexpected lag
+	return keepAliveBytes + handshakeBytes
 }

client/internal/lazyconn/inactivity/manager_test.go

@@ -25,7 +25,7 @@ func TestNewManager(t *testing.T) {
 
 		t.Run(fmt.Sprintf("Scenario %d", i), func(t *testing.T) {
 			mock := newMockWgInterface("peer1", sc.Data, timer)
-			manager := NewManager(mock)
+			manager := NewManager(mock, nil)
 			peerCfg := &lazyconn.PeerConfig{
 				PublicKey: "peer1",
 				Log:       log.WithField("peer", "peer1"),

client/internal/lazyconn/inactivity/recorder.go

@@ -0,0 +1,58 @@
+package inactivity
+
+import (
+	"fmt"
+	"os"
+	"sync"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+)
+
+type Recorder struct {
+	mu       sync.Mutex
+	file     *os.File
+	filename string
+}
+
+func NewRecorder() *Recorder {
+	timestamp := time.Now().Format("20060102_150405")
+	filename := fmt.Sprintf("inactivity_log_%s.txt", timestamp)
+	file, err := os.OpenFile(filename, os.O_CREATE|os.O_WRONLY|os.O_APPEND, 0644)
+	if err != nil {
+		log.Errorf("error opening file: %v", err)
+	}
+
+	return &Recorder{
+		file:     file,
+		filename: filename,
+	}
+}
+
+func (r *Recorder) ReceivedBytes(peer string, now time.Time, bytes int64) {
+	if r == nil {
+		return
+	}
+
+	r.mu.Lock()
+	defer r.mu.Unlock()
+
+	entry := fmt.Sprintf("%s; %s; %d\n", now.Format(time.RFC3339), peer, bytes)
+	_, err := r.file.WriteString(entry)
+	if err != nil {
+		log.Errorf("error writing to file: %v", err)
+	}
+}
+
+func (r *Recorder) Close() {
+	if r == nil {
+		return
+	}
+
+	r.mu.Lock()
+	defer r.mu.Unlock()
+
+	if err := r.file.Close(); err != nil {
+		log.Errorf("error closing file: %v", err)
+	}
+}

client/internal/lazyconn/inactivity/scenarios_test.go

@@ -52,52 +52,60 @@ var scenarios = []scenario{
 	{
 		ExpectedInactive: true,
 		Data: []rxHistory{
+			//96
 			{when: 0 * time.Second, RxBytes: 32},
 			{when: 25 * time.Second, RxBytes: 32},
 			{when: 50 * time.Second, RxBytes: 32},
+
+			//212
 			{when: 75 * time.Second, RxBytes: 32},
 			{when: 100 * time.Second, RxBytes: 32},
 			{when: 100 * time.Second, RxBytes: 148},
+
+			//96
 			{when: 125 * time.Second, RxBytes: 32},
 			{when: 150 * time.Second, RxBytes: 32},
 			{when: 175 * time.Second, RxBytes: 32},
+
+			//212
 			{when: 200 * time.Second, RxBytes: 32},
 			{when: 225 * time.Second, RxBytes: 32},
 			{when: 225 * time.Second, RxBytes: 148},
+
+			//96
 			{when: 250 * time.Second, RxBytes: 32},
 			{when: 275 * time.Second, RxBytes: 32},
 			{when: 300 * time.Second, RxBytes: 32},
+
 			{when: 325 * time.Second, RxBytes: 32},
 			{when: 350 * time.Second, RxBytes: 32},
 			{when: 350 * time.Second, RxBytes: 148},
+
 			{when: 375 * time.Second, RxBytes: 32},
 			{when: 400 * time.Second, RxBytes: 32},
+
 			{when: 425 * time.Second, RxBytes: 32},
 			{when: 450 * time.Second, RxBytes: 32},
 			{when: 475 * time.Second, RxBytes: 32},
 			{when: 475 * time.Second, RxBytes: 148},
+
 			{when: 500 * time.Second, RxBytes: 32},
 			{when: 525 * time.Second, RxBytes: 32},
+
 			{when: 550 * time.Second, RxBytes: 32},
 			{when: 575 * time.Second, RxBytes: 32},
 			{when: 600 * time.Second, RxBytes: 32},
 			{when: 600 * time.Second, RxBytes: 148},
+
 			{when: 625 * time.Second, RxBytes: 32},
 			{when: 650 * time.Second, RxBytes: 32},
+
 			{when: 675 * time.Second, RxBytes: 32},
 			{when: 700 * time.Second, RxBytes: 32},
+
 			{when: 725 * time.Second, RxBytes: 32},
 			{when: 725 * time.Second, RxBytes: 148},
 			{when: 750 * time.Second, RxBytes: 32},
 		},
 	},
-	{
-		ExpectedInactive: false,
-		Data: []rxHistory{
-			{when: 0 * time.Second, RxBytes: 32},
-			{when: 25 * time.Second, RxBytes: 32},
-			{when: 50 * time.Second, RxBytes: 100},
-			{when: 75 * time.Second, RxBytes: 32},
-		},
-	},
 }

client/internal/lazyconn/manager/manager.go

@@ -67,6 +67,7 @@ type Manager struct {
 // engineCtx is the context for creating peer Connection
 func NewManager(config Config, engineCtx context.Context, peerStore *peerstore.Store, wgIface lazyconn.WGIface, connStateDispatcher *dispatcher.ConnectionDispatcher) *Manager {
 	log.Infof("setup lazy connection service")
+
 	m := &Manager{
 		engineCtx:            engineCtx,
 		peerStore:            peerStore,
@@ -76,19 +77,11 @@ func NewManager(config Config, engineCtx context.Context, peerStore *peerstore.S
 		managedPeersByConnID: make(map[peerid.ConnID]*managedPeer),
 		excludes:             make(map[string]lazyconn.PeerConfig),
 		activityManager:      activity.NewManager(wgIface),
-		inactivityManager:    inactivity.NewManager(wgIface),
+		inactivityManager:    inactivity.NewManager(wgIface, config.InactivityThreshold),
 		peerToHAGroups:       make(map[string][]route.HAUniqueID),
 		haGroupToPeers:       make(map[route.HAUniqueID][]string),
 	}
 
-	if config.InactivityThreshold != nil {
-		if *config.InactivityThreshold >= inactivity.MinimumInactivityThreshold {
-			m.inactivityThreshold = *config.InactivityThreshold
-		} else {
-			log.Warnf("inactivity threshold is too low, using %v", m.inactivityThreshold)
-		}
-	}
-
 	m.connStateListener = &dispatcher.ConnectionListener{
 		OnConnected:    m.onPeerConnected,
 		OnDisconnected: m.onPeerDisconnected,
@@ -620,5 +613,5 @@ func (m *Manager) onPeerDisconnected(peerConnID peerid.ConnID) {
 	}
 
 	// todo reset inactivity monitor
-	mp.peerCfg.Log.Infof("--- peer disconnected, stopping inactivity monitor?")
+	mp.peerCfg.Log.Warnf("--- peer disconnected, stopping inactivity monitor?")
 }